Correction

NEUROSCIENCE Correction for “Genetic rescue models refute nonautonomous in issue 20, May 16, 2017, of Proc Natl Acad Sci USA (114:5259– death in ,” by Susanne F. Koch, 5264; first published May 3, 2017; 10.1073/pnas.1615394114). Jimmy K. Duong, Chun-Wei Hsu, Yi-Ting Tsai, Chyuan-Sheng Lin, The authors note that Fig. 3 appeared incorrectly. The cor- Christian A. Wahl-Schott, and Stephen H. Tsang, which appeared rected figure and its legend appear below.

Fig. 3. Restoration of PDE6b expression rescues rods and partially restores OS length, in a dose-dependent fashion, and prevents cone death. Pde6bSTOP/ Pde6bH620Q, Pde6g::CreERT2 mice were injected with 25, 50, or 100 μg/g BW tamoxifen to rescue 30% (T30), 50% (T50), or 70% (T70) of rods, respectively; + untreated mutants (mut) and wild-type mice (Pde6b /Pde6bH620Q, Pde6g::CreERT2, WT) were not injected. Tamoxifen injections were in 1-mo-old mice; at 9 mo of age, retinas were sectioned and immunolabeled. (A) Antibodies: or PDE6b (rod OSs, red) and arrestin (cones, green). Hoechst dye, nuclei, blue. Arrows, arrestin-immunopositive cone cell bodies; arrowheads, arrestin-immunopositive cone OSs. (B) Rhodopsin and arrestin immunolabeled sections were quantitatively analyzed for ONL thickness, cone number, rod OS length and cone IS + OS length. Gray dots, individual mice (n = 4 for each group); horizontal black lines, group means. Treated and untreated groups were compared by a linear regression model: ***P < 0.001. IS, inner segment; ONL, outer nuclear layer; OS, outer segment. (Scale bar, 15 μm.)

www.pnas.org/cgi/doi/10.1073/pnas.1708940114

E5484 | PNAS | July 3, 2017 | vol. 114 | no. 27 www.pnas.org Downloaded by guest on September 30, 2021 Genetic rescue models refute nonautonomous rod cell death in retinitis pigmentosa

Susanne F. Kocha,b,c, Jimmy K. Duongd,Chun-WeiHsua,b,c, Yi-Ting Tsaia,b,c,e, Chyuan-Sheng Linf, Christian A. Wahl-Schottg, and Stephen H. Tsanga,b,c,d,f,1 aJonas Children’s Vision Care, Departments of Ophthalmology, Pathology, and Cell Biology, Columbia University, New York, NY 10032; bBernard & Shirlee Brown Glaucoma Laboratory, Departments of Ophthalmology, Pathology, and Cell Biology, Columbia University, New York, NY 10032; cEdward S. Harkness Eye Institute, New York Presbyterian Hospital, New York, NY 10032; dDepartment of Biostatistics, Mailman School of Public Health, Columbia University Medical Center, New York, NY 10032; eInstitute of Human Nutrition, College of Physicians and Surgeons, Columbia University, New York, NY 10032; fHerbert Irving Comprehensive Cancer Center, Columbia University Medical Center, New York, NY 10032; and gCenter for Integrated Protein Science Munich (CIPSM)attheDepartmentofPharmacy–Center for Drug Research, Ludwig-Maximilians-Universität München, 81377 Munich, Germany

Edited by Roderick R. McInnes, Lady Davis Institute, McGill University, Montreal, Canada, and accepted by Editorial Board Member Jeremy Nathans April 7, 2017 (received for review September 14, 2016)

Retinitis pigmentosa (RP) is an inherited neurodegenerative Results disease, in which the death of mutant rod photoreceptors leads Cre-Driven Rescue Produces Mosaic Retinas in Which the Ratio secondarily to the non-cell autonomous death of cone photore- of Mutant and Wild-Type Photoreceptors Is Controlled. To in- ceptors. Gene therapy is a promising treatment strategy. Un- vestigate the fate of rescued rods in a pathological environment, fortunately, current methods of gene delivery treat only a we used two different strategies. For both, we created RP mice fraction of diseased cells, yielding retinas that are a mosaic of whose retinal photoreceptor layer is a mosaic of treated and treated and untreated rods, as well as cones. In this study, we untreated rods—as well as cones. In one strategy, treated and created two RP mouse models to test whether dying, untreated untreated rods were spatially segregated; in the second, they rods negatively impact treated, rescued rods. In one model, were diffusely intermixed, and the percentage of rescued rods treated and untreated rods were segregated. In the second was controlled. For both models, we used our genetically engi- model, treated and untreated rods were diffusely intermixed, neered RP mouse model Pde6bSTOP/Pde6bH620Q, in which one and their ratio was controlled to achieve low-, medium-, or high- efficiency rescue. Analysis of these mosaic retinas demonstrated allele of rod-specific Pde6b contains a point mutation and the that rescued rods (and cones) survive, even when they are second allele a floxed STOP cassette. In these mice, PDE6b is dramatically reduced, leading to rod death and secondary de- greatly outnumbered by dying photoreceptors. On the other STOP H620Q hand, the rescued photoreceptors did exhibit long-term defects generation of cones (4). When Pde6b /Pde6b mice are in their outer segments (OSs), which were less severe when more crossed with a Cre transgenic line, the STOP cassette is removed photoreceptors were treated. In summary, our study suggests and PDE6b is expressed in cells where Cre is expressed (5). We that even low-efficiency gene therapy may achieve stable used two different Cre drivers to control the pattern and/or NEUROSCIENCE survival of rescued photoreceptors in RP patients, albeit with number of rescued rods. The first driver, Pax6α::Cre, encodes OS dysgenesis. Cre recombinase under the control of a retina-specific regulatory element (α) of murine Pax6, a transcription factor expressed in neurodegeneration | retinitis pigmentosa | death | retinal progenitor cells that gives rise to cells in distal retina (6). non-cell autonomous degeneration | gene therapy The second driver, Pde6g::CreERT2, is tamoxifen-inducible and encodes Cre recombinase under the control of the rod-specific etinitis pigmentosa (RP) is a group of retinal degenerative Rdiseases and the most common cause of inherited blindness Significance (1). Most often, RP results from mutations in rod-specific , which trigger the cell-autonomous loss of rods that, in turn, Retinitis pigmentosa is the leading cause of inherited blind- causes the non-cell autonomous loss of cones (2). Gene therapy ness. Although gene therapy has the capacity to rescue dis- strategies are being intensively developed and tested for RP and eased cells (usually rods), current methods generate retinas other inherited retinal degenerative diseases. However, the that are a mix of treated, rescued and untreated, dying rods. To barriers to developing a successful gene therapy are significant. determine whether the dying rods negatively impact rescue, For example, current methods of gene delivery (in both humans we developed mouse models that allowed us to treat defined and mice) transduce only a fraction of the diseased cells. In fractions of diseased rods. We found that dying rods did not retinas, the result is a mosaic of treated/rescued rods (and trigger the death of rescued photoreceptors, even when the cones), surrounded by large numbers of untreated/diseased rods. rescued cells are greatly outnumbered. On the other hand, the Here, we tested whether untreated dying rods impact survival, rescued photoreceptors did exhibit long-term defects, which structure, and/or function of treated rods. In addition, we set out were less severe when more rods were treated. Thus, although to examine the non-cell autonomous effects of dying rods on genetic rescue leads to survival of treated rods, it does not cones in greater detail. To do this, we generated mice in which prevent other aspects of the retinitis pigmentosa pathology. the photoreceptor layer is a mosaic of treated and untreated mutant rods and rescue is spatially or numerically controlled. In Author contributions: S.F.K. and S.H.T. designed research; S.F.K., C.-W.H., Y.-T.T., and C.-S.L. performed research; S.F.K., J.K.D., C.A.W.-S., and S.H.T. analyzed data; and S.F.K. these mice, the mutant rods lack the gene encoding rod-specific and S.H.T. wrote the paper. — cGMP phosphodiesterase 6b (Pde6b) a common cause of au- The authors declare no conflict of interest. tosomal recessive RP (3). Analysis of our mosaic retinas revealed This article is a PNAS Direct Submission. R.R.M. is a guest editor invited by the Editorial that untreated dying rods did not impact survival of rescued rods Board. or cones but did trigger outer segment (OS) dysgenesis in nearby 1To whom correspondence should be addressed. Email: [email protected]. rods and cones, which persisted after the mutant rods have This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. all died. 1073/pnas.1615394114/-/DCSupplemental. www.pnas.org/cgi/doi/10.1073/pnas.1615394114 PNAS | May 16, 2017 | vol. 114 | no. 20 | 5259–5264 Rescued Photoreceptors Appear to Migrate into Untreated Zones and Exhibit OS Shortening That Is More Dramatic the Farther the Migration. ROSAnT-nG, Pax6α::Cre reporter mice have retinas with large, roughly segregated zones of recombined and non- recombined cells (Fig. 1A). A single subretinal injection of gene therapy vectors (in both animals and humans) tends to yield a similar pattern of rescued and nonrescued cell zones. We therefore crossed Pde6bSTOP/Pde6bH620Q and Pax6α::Cre mice to generate retinas with the same pattern seen in Fig. 1A and used PDE6b immunolabeling (rod OSs) to visualize areas with res- cued rods. In retinas from 11-mo-old Pde6bSTOP/Pde6bH620Q, Pax6α::Cre mice, immunolabeling was absent around the optic nerve—extending 138 ± 35 μm and 224 ± 47 μm in the superior and inferior directions, respectively (n = 3 retinas) (Fig. 2A); that is, the immunonegative zone was much smaller than the red nonrecombined zone (see Fig. 1A and associated text). Thus, at 11 mo, PDE6b expression was much closer to the optic nerve, suggesting that rescued rods may migrate into the untreated Fig. 1. Strategies for generating spatially and numerically reproducible zone, but this hypothesis remains to be investigated. To assess mosaic patterns of rod rescue using two different Cre drivers. (A) Two-color STOP H620Q α — nT-nG retinal morphology in Pde6b /Pde6b , Pax6 ::Cre mice fluorescent ROSA , Pax6α::Cre mice generated by crossing Pax6α-Cre and + H620Q nT-nG in comparison with wild-type (Pde6b /Pde6b ) and mutant ROSA -reporter mice. Representative composite image of a retina from STOP H620Q — a 3-mo-old mouse, sectioned through the ON and immunolabeled for (Pde6b /Pde6b ) mice retinal sections from 11-mo-old GFP (green, Cre-recombined cells). Arrowheads demark the border between mice were stained with PDE6b antibody (rod OSs), PNA lectin predominantly nonrecombined and recombined areas. (B) Two-color fluo- (cone OS—as well as cone synapses), and Hoechst dye (nuclei) rescent ROSAnT-nG, Pde6gCre::ERT2 mice generated by crossing Pde6g:: (Fig. S3A). In mutant retinas, ONL was almost gone due to pho- CreERT2 and ROSAnT-nG-reporter mice; 1-mo-old mice were injected with 25, toreceptor cell loss, and PDE6b-labeled rod OSs and PNA- 50, or 100 μg/g BW tamoxifen; 2 wk later, retinas were sectioned and GFP STOP H620Q × labeled cone OSs were completely absent. In Pde6b /Pde6b , immunolabeled. (C) Number of green and red nuclei was counted in a 90 Pax6α::Cre retinas, untreated, PDE6b-negative areas were in- 90 μm area. y axis, total green nuclei divided by red + green nuclei; gray dots, individual mice (n values indicated on x axis); horizontal black lines, distinguishable from mutant retinas, except that the former group means; red, tomato-fluorescence (nonrecombined cells). INL, inner exhibited some PNA staining; treated areas were indistinguishable nuclear layer; ON, optic nerve; ONL, outer nuclear layer. [Scale bars, 100 μm from wild-type retinas. In the most central part of the PDE6b- (A) and 15 μm(B).] positive zone (i.e., bordering the PDE6b-negative zone), ONL promoter of Pde6g, which encodes the gamma subunit of rod cGMP phosphodiesterase (5). To define the patterns of Pax6α::Cre and Pde6g::CreERT2 recombinase activity, we crossed our Cre transgenic lines with the ROSAnT-nG reporter mouse line. ROSAnT-nG mice contain a fluorescent allele that expresses nuclear-localized red fluores- cence (tdTomato) in all cells, except those exposed to Cre recombinase; the latter cells express nuclear-localized enhanced green fluorescent protein (EGFP) (7). In ROSAnT-nG, Pax6α::Cre flat-mounted retinas and sections from 3-mo-old mice, Cre- recombined cells (green) were detected predominantly in the distal retina (Fig. S1A and Fig. 1A); in flat-mounted retinas (n = 9), the green, Cre-recombined area was about 63 ± 1.3% (Fig. S1B), and in sections (n = 4 retinas) it started around 415 ± 30 μm (superior) and 675 ± 51 μm (inferior) from the optic nerve (Fig. 1A). Because the border between recombined and non- recombined areas is not sharp (Fig. 1A, arrowheads), we defined it as the point at which the overwhelming majority of cells lo- nT-nG cated more centrally were red (nonrecombined). In ROSA , Fig. 2. Sustained shortening of rod and cone OSs, even after the period of Pde6g::CreERT2 mice, red and green nuclei were uniformly rod cell death. (A) Retinal section (composite) from an 11-mo-old Pde6bSTOP/ distributed over the entire photoreceptor cell body-containing Pde6bH620Q, Pax6α::Cre mouse immunolabeled for PDE6b (red, rod OSs) and outer nuclear layer (ONL). As expected, only red nuclei were stained with Hoechst dye (blue, nuclei). (B and C) Rod OS lengths and cone IS + OS lengths were measured in 11-mo-old retinal sections from wild-type (red found in the inner nuclear layer (INL), which does not contain STOP H620Q α — circles), mutant (gray diamonds), and Pde6b /Pde6b , Pax6 ::Cre mice photoreceptor cell bodies (Fig. 1B) and all cone nuclei were (blue triangles). Treated zones, green bars; untreated zones, white and gray red (Fig. S2), because the Pde6g::CreERT2 driver is only active in bars. The border between treated and untreated zones (arrowheads here rods. The mean percentages of green Cre-recombined rod nuclei and in Fig. 1A) was measured in sections from ROSAnT-nG, Pax6α::Cre mice in the ROSAnT-nG, Pde6g::CreERT2 mice injected with tamoxifen (as in Fig. 1A; n = 4 mice). PDE6b-negative zone (gray bar) was measured in STOP H620Q at 25, 50, or 100 μg/g body weight (BW) were 31% ± 6, 46% ± 6, 11-mo-old sections from Pde6b /Pde6b , Pax6α::Cre mice (as in A). – and 72% ± 2, respectively (i.e., roughly 30%, 50%, and 70%) Arrows in A C demark the outer, peripheral border of the PDE6b-negative zone; bold symbols, group means; lighter colored symbols, individual mice (Fig. 1C). Thus, Pde6g::CreERT2 activity can be controlled (n = 3, for every group). Each line connects the group means. Black vertical in vivo, albeit with variation, to achieve defined percentages of line, optic nerve. IS, inner segment; ON, optic nerve; ONL, outer nuclear layer; rods with allelic conversion. OS, outer segment. (Scale bar, 100 μm.)

5260 | www.pnas.org/cgi/doi/10.1073/pnas.1615394114 Koch et al. thickness and OS lengths were graded. We next quantified ONL To mimic RP gene therapy at midstage disease, 1-mo-old thickness, cone number, rod OS length, and cone IS + OS length at Pde6bSTOP/Pde6bH620Q, Pde6g::CreERT2 mice, which have lost multiple positions across a 1,500-μm length of the central retina roughly 30% of their rods (5), were injected with 25, 50, or 100 μg/g (less than half of the area shown in Fig. 2A); identical quantifi- BW tamoxifen (i.e., low-, medium-, and high-efficiency genetic cations were performed in wild-type and mutant retinas. In the rescue). We demonstrated (Fig. 1C) that these three doses rescue PDE6b-negative area (gray bar in Fig. 2 B and C and Fig. S3 B and ∼30%, 50%, and 70% of the remaining rods. Therefore, our C), ONL thickness, rod OS length, and cone IS + OS length were treatment rescued ∼20%, 35%, or 50% of the total, original rods. not significantly different from mutant retinas; however, cone To analyze the effects of this treatment on retinal structure, retinas number was significantly increased (P < 0.01). In the area just from 9-mo-old mice were sectioned and immunolabeled. To dis- outsidethePDE6b-negativezone(whitebarsinFig.2B and C and tinguish rods and cones, we immunolabeled sections with rho- Fig. S3 B and C),ONLthicknessandOSlengthweregraded dopsin antibody (rod OSs) and arrestin antibody (cones) (Fig. 3A, (values decreased as the distance to the PDE6b-negative zone Top). In treated retinas, there was clear dose-dependent rescue of decreased). The graded ONL appears to be mainly the result of rhodopsin-positive rod OSs and the ONL. In addition, the arrestin immunolabeling revealed dramatic dose-dependent rescue of treated photoreceptors moving into untreated areas, but there may — be alternative explanations (Fig. S3 B and C;notebluetrianglesin cones especially cone OS length. The small amount of arrestin immunolabeling in mutant retinas represents the fraction of cone white bars). On the other hand, the graded OS lengths in these cells—devoid of their OSs—that persist for some time after mutant 11-mo retinas (by which time, diseased rods have all died) suggest rods have died (8). that some unknown pathological feature(s), other than dying rods, The OSs of treated retinas exhibited PDE6b immunolabeling sustains the shortened OS phenotype. Finally, in treated areas (Fig. 3A, Bottom)—thereby demonstrating survival of treated (green bars in Fig. 2 B and C and Fig. S3 B and C), all four features rods. Further, we performed PCR analysis on DNA isolated were not significantly different from wild-type retinas, demon- from the retinal ONL. A 415-bp band was found in DNA am- strating that rescued rods and cones were not negatively impacted plified from mutant mice but not in DNA amplified from treated by distant degenerating rods. On the other hand, the increased mutant ONL (Fig. S5). The removal of the STOP cassette in numbers of cones in untreated areas might result from migration treated rods leads to a shorter 362-bp band; therefore, all rods in of rescued cones from treated areas (Discussion). the treated retinas were devoid of the STOP cassette. To quantify photoreceptor rescue, we measured ONL thick- Treated Rods, Cones, and Their OSs Are All Rescued in a Dose- ness (rods + cones), cone numbers, rod OS length, and cone IS + Dependent Fashion. nT-nG Using our ROSA , Pde6g::CreERT2 re- OS length (Fig. 3B). All four structural measures were signifi- porter mice, we demonstrated that the percentage of recombined cantly greater in treated retinas, compared with untreated mu- rods can be controlled by the tamoxifen dose (Fig. 1 B and C). We tants (P < 0.001). ONL thickness, rod OS length, and cone IS + STOP H620Q therefore crossed Pde6b /Pde6b and Pde6g::CreERT2 OS length were all significantly less than wild type (P < 0.001). mice to generate RP mice with the same degree of dose–response ONL thickness was less than wild type because treatment was control. To verify that PDE6b expression correlates with tamoxi- administered at 1 mo, when 30% of rods have already died, and STOP H620Q fen dose, we injected tamoxifen in 19-d-old Pde6b /Pde6b , because only 70% or less of the remaining rods were treated. Pde6g::CreERT2 mice (i.e., before photoreceptor degeneration) Similarly, it is not surprising that rod OS and cone IS + OS NEUROSCIENCE (5) and then performed immunoblot analysis on retinas when the lengths were not fully restored to wild-type lengths by 9 mo, mice were 12 wk old. PDE6b expression was, in fact, proportional given that the rescued photoreceptors inhabit a retina that has to tamoxifen dose (Fig. S4). lost 50–80% of the original rods. In addition, rescue of ONL, rod

Fig. 3. Restoration of PDE6b expression rescues rods and partially restores OS length, in a dose-dependent fashion, and prevents cone death. Pde6bSTOP/ Pde6bH620Q, Pde6g::CreERT2 mice were injected with 25, 50, or 100 μg/g BW tamoxifen to rescue 30% (T30), 50% (T50), or 70% (T70) of rods, respectively; untreated mutants (mut) and wild-type mice (Pde6b+/Pde6bH620Q, Pde6g::CreERT2, WT) were not injected. Tamoxifen injections were in 1-mo-old mice; at 9 mo of age, retinas were sectioned and immunolabeled. (A) Antibodies: rhodopsin or PDE6b (rod OSs, red) and arrestin (cones, green). Hoechst dye, nuclei, blue. Arrows, arrestin-immunopositive cone cell bodies; arrowheads, arrestin-immunopositive cone OSs. (B) Rhodopsin and arrestin immunolabeled sections were quantitatively analyzed for ONL thickness, cone number, rod OS length and cone IS + OS length. Gray dots, individual mice (n = 4 for each group); horizontal black lines, group means. Treated and untreated groups were compared by a linear regression model: ***P < 0.001. IS, inner segment; ONL, outer nuclear layer; OS, outer segment. (Scale bar, 15 μm.)

Koch et al. PNAS | May 16, 2017 | vol. 114 | no. 20 | 5261 phenotype; although all retinas were processed at the same time, only the T50 retinas displayed the pronounced abnormality.

Rod and Cone Function Are Rescued, and That Rescue Is Dose- Dependent. To determine if retinal function was rescued, 1-mo-old Pde6bSTOP/Pde6bH620Q, Pde6g::CreERT2 mice were tamoxifen injected (or not), and 8 mo later, electroretinography (ERG) analysis was performed. To derive mixed rod and cone ERG responses, mice were dark-adapted and then eyes were exposed to bright flashes. Wild-type retinas exhibited a photoreceptor- mediated a-wave (negative deflection) followed by an inner retina-mediated b-wave (positive deflection). In contrast, un- treated mutant retinas lacked the a-wave and exhibited a strongly reduced b-wave. Treatment rescued the a- and b-waves (Fig. 5 A and B); as expected, rescue was not to wild-type levels, as treatment was administered at 1 mo, when about 30% of rods had already died and 70% or fewer of the remaining rods were treated (i.e., ∼50% or fewer of the total/original rods). In mutant retinas in which 50% and 70% of the remaining rods were rescued (and shown in Fig. 3C to be significantly differ- ent), functional rescue was not significantly different (a-waves, P = 0.8; b-waves, P = 0.7). This suggests that the relationship between structural and functional rescue is nonlinear and/or that the variability in a- and b-wave amplitudes may decrease the ability to detect subtle differences. In these T50 and T70 retinas, a- and b-wave amplitudes were significantly greater than either untreated mutant retinas (P < 0.001) or T30 retinas (T50, P < 0.003 and 0.002, respectively; T70, P < 0.001). In T30 retinas, b-wave amplitudes were significantly greater than untreated mutants (P < 0.001), but a-wave amplitudes were not (P = 0.1).

Analysis of Disease Progression in Mosaic Retinas Suggests Non-Cell Autonomous Triggers of OS Dystrophy and Cell-Autonomous Triggers Fig. 4. Ultrastructural dysgenesis in OSs of rescued retinas. Pde6bSTOP/ of Rod Death. To understand disease progression in our rescued Pde6bH620Q, Pde6g::CreERT2 mice were injected with 50 μg/g BW tamoxifen STOP H620Q + mosaic retinas, 1-mo-old Pde6b /Pde6b , Pde6g-CreERT2 to rescue 50% (T50) of rods, respectively; wild-type mice (Pde6b /Pde6bH620Q, Pde6g::CreERT2, WT) were not injected. Tamoxifen injections were in 1-mo-old mice; at 8 mo of age, retinas were sectioned and processed for electron micros- copy. Wt is a composite image. Arrows, normal OSs; arrowheads, disrupted OSs; vertical black bars, approximate OS length. IS, inner segment; ONL, outer nuclear layer; OS, outer segment. RPE, retinal pigment epithelium. (Scale bar, 2 μm.)

OS, and (for the most part) cone IS + OS length were dose- dependent: 70% rod-rescued retinas vs. 50% or 30% (P < 0.001); ONL thickness and rod OS length in T50 vs. T30 (P < 0.003 and P < 0.008, respectively). Importantly, in all treated retinas, even in those in which only 30% of remaining rods were rescued, cone numbers were not significantly different from wild type (P = 0.7, 0.9 and 0.9, respectively). In summary, our gene-rescue strategy in midstage disease mice led to the dose-dependent structural rescue of rods and prevented non-cell autonomous cone death.

Rescued OSs Are Dysmorphic—Even After the Phase of Rod Death. To further assess structural changes in OSs, 1-mo-old Pde6bSTOP/ Pde6bH620Q, Pde6g::CreERT2 mice were tamoxifen-injected (or not), and at 8 mo, retinas were analyzed by light and electron microscopy. Light microscopy of H&E stained sections (Fig. S6A) showed normal OSs in wild-type retinas, no OSs in untreated mutants, and rescued OSs in treated retinas. As shown in Fig. 3, Fig. 5. Rescue of photoreceptor function is proportional to the number of rescue of the OS length was clearly dose-dependent and partial. treated rods. Pde6bSTOP/Pde6bH620Q, Pde6g::CreERT2 mice were injected The dose-dependent shortening of OSs was confirmed in electron with 25, 50, or 100 μg/g BW tamoxifen to rescue 30% (T30), 50% (T50), or micrographs of treated retinas (Fig. 4 and Fig. S6B). In addition, in 70% (T70) of rods, respectively; untreated mutants (mut) and wild-type mice + H620Q the T50 retinas, we noted holes in the OS layer and that the OSs (Pde6b /Pde6b , Pde6g::CreERT2, WT) were not injected. Tamoxifen in- were no longer arranged in cohesive linear arrays. Further, in some jections were in 1-mo-old mice; at 9 mo of age, retinal function was analyzed by ERG. (A) Representative ERG responses. (B) Quantification of ERG a-wave OSs, the membranous discs appeared disrupted (Fig. 4, detail). and b-wave amplitudes. Gray dots, individual eyes (n values indicated on x Although such changes are sometimes seen in wild-type EM im- axis); horizontal black lines, group means. Treated and untreated groups ages, we noted a clear dose dependence to the ultrastructural were compared using a linear mixed model: ***P < 0.001.

5262 | www.pnas.org/cgi/doi/10.1073/pnas.1615394114 Koch et al. mice were tamoxifen injected (or not), and photoreceptor layer turn, implies that OS shortening is driven by non-cell autonomous structural features were measured at 1, 2.5, 5, and 9 mo of age mechanisms, triggered by dying untreated rods. (Fig. 6 and Fig. S7). In wild-type retinas, photoreceptor layer Discussion features were stable over the 8-mo study: ONL thickness, P = 0.9; rod OS length, P = 0.5; cone IS + OS length, P = 0.5 (P In RP, degeneration and, subsequently, death of diseased, mu- values are for 1 vs. 9 mo). In contrast, untreated, mutant retinas tant rods leads to the non-cell autonomous loss of wild-type exhibited progressive and dramatic thinning of the ONL (P < cones (2, 10, 11). However, it is not known if degenerating rods 0.001) and shortening of both rod and cone OSs (P < 0.001). have non-cell autonomous effects on the rods surrounding them. — T30 and T50 retinas also exhibited significant decreases in ONL This issue is most relevant in gene therapy-treated retinas in thickness (P < 0.006) but only over the first 4 mo; there was no terms of understanding whether rescued rods are negatively significant change over the last 4 mo (P = 0.9 and 0.5 for T30 and impacted by untreated rods. To study this, we used two different T50, respectively) (Fig. 6). T70 retinas exhibited a modest de- Cre drivers to express a therapeutic gene in diseased rods in a mouse model of RP. One driver, Pax6α::Cre, yielded retinas in crease in ONL thickness over the first 4 mo that was not sig- which treated and untreated rods were roughly segregated, nificant (P = 0.09). Thus, in treated mutants, the only significant thereby modeling human gene therapy-treated retinas. The decrease in ONL thickness was in retinas from mice treated at second driver, Pde6g::CreERT2, generated retinas in which res- the two lower doses and only for the first 4 mo following treat- cued rods were diffusely distributed over the entire photore- ment. If this decrease is due entirely to cell-autonomous death of ceptor layer (i.e., treated and untreated rods were intermixed), untreated rods, then we would predict stable survival of all of the and importantly, the percentage of rescued rods was controlled. treated rods (i.e., 31%, 46%, and 72% of remaining rods for the In both mice, we did not restore the mutant gene in all diseased three tamoxifen doses, respectively; Fig. 1C). In fact, at 9 mo, rods, as we have done before (5). Data from these two distinct actual rod survival values were 31%, 46%, and 82%, which are recombinant retinas strongly suggest that rod rescue is stable, not significantly different from the predicted values (P = 0.6, 0.9, even when the vast majority of rods are untreated (i.e., mutant). and 0.3, respectively). Cones were not taken into account, as they The findings from this study and a previous one (5) support two make up only 3% of photoreceptors in normal retinas (9). Thus, of the major predictions of the “one-hit” model of cell death these data provide indirect evidence for the stable, sustained (12): (i) rod photoreceptor death is cell-autonomous (this study), survival of the vast majority of (if not all) treated rods in a dis- and (ii) mutant photoreceptors can be rescued even when the eased environment. We found that the overall dose-dependent activity of the relevant gene is restored at late-stage disease (5). pattern of OS shortening over the 8-mo timeframe was the same Although our quantitative analysis suggests that few, if any, as ONL thickness (Fig. S7 and Fig. 6, respectively). For T30 and rescued rods died, we did observe nonlethal phenotypes in sur- T50 retinas, rod and cone OS length decreased significantly over viving photoreceptors. Specifically, where dying and rescued the first 4 mo (P < 0.05) but not over the last 4 (Fig. S7 A and B, photoreceptors were intermixed, rod and cone OSs were short- respectively). For T70 retinas, there was no significant change in ened. Although the highest efficiency treatment (70% rescue of H620Q either rod OS or cone IS + OS lengths over the 8-mo period (P = remaining rods and 50% rescue of total rods) in Pde6b / STOP 0.7 and 0.5, respectively). These data suggest that in treated Pde6b , Pde6g::CreERT2 mice halted the shortening, OSs in retinas, rod and cone OSs progressively shorten until most, if not the lower efficiency rescued retinas (50% and 30% rescue of NEUROSCIENCE all, untreated mutant rods have died (around 5 mo). This, in remaining rods and 35% and 21% of total rods, respectively) continued to shorten but then plateaued at the same time as ONL thickness (i.e., when most untreated rods have died). These dose–response disease-progression data suggest that OS short- ening is triggered non-cell autonomously by dying mutant rods, and only manifests when there is a preponderance of dying rods. Indeed, OS shortening was not apparent within rescued zones in our Pax6α::Cre retinas. Thus, OS shortening might not occur within rescued regions of treated human RP retina. The finding that shortened OSs do not regrow to their normal length indicates a shift in the homeostasis between OS mor- phogenesis and retinal pigment epithelial (RPE) cell phagocy- tosis (13, 14). Under favorable conditions, rod and cone OSs are capable of regrowth—for example, following light-induced damage (15), retinal detachment (16), or macular hole surgery (17). However, the extent and time course of regrowth depend on the severity of the damage (15). For example, retinal genetic rescue at advanced disease stages does not lead to OS re- generation (5, 18, 19), suggesting a correlation between photo- receptor numbers and OS synthesis. Perhaps a critical number of Fig. 6. Dying rods do not diminish the long-term survival of neighboring rods might be needed to release glucose from RPE, where it is rescued rods. Pde6bSTOP/Pde6bH620Q, Pde6g::CreERT2 mice were injected sequestered following rod degeneration and thus unavailable to with 25, 50, or 100 μg/g BW tamoxifen to rescue 30% (T30), 50% (T50), or support OS synthesis (8, 20). Interestingly, in untreated RP pa- 70% (T70) of rods, respectively; untreated mutants (mut) and wild-type mice tients (and mice), the earliest detected histopathologic change is + (Pde6b /Pde6bH620Q, Pde6g::CreERT2, WT) were not injected. Tamoxifen in- rod OS dysgenesis. Our data demonstrate that OS dysgenesis jections were in 1-mo-old mice. Wild-type and untreated mutant mice were does not, by itself, necessarily trigger cell death. This suggests killed at 1, 2.5, 5, and 9 mo; treated mice were killed at 2.5, 5, and 9 mo that OS dysgenesis and cell death are phenotypes of two separate (9-mo data are also shown in Fig. 3). Each symbol represents an individual mouse; n = 4 for all genotypes and/or treatments and time points, except n = pathogenesis pathways. 3 for untreated mutants at 1 mo and for all groups at 2.5 mo. Mutant mice Our experiments, which were designed to study the fate of were fitted to an exponential decay model; treated mice were fitted to an rescued rods in a diseased environment, strongly suggest that exponential decay model with plateau (see Eq. S1 in SI Materials and survival of rescued rods is not affected by dying, untreated rods. Methods). Two previous mouse studies came to the opposite conclusion

Koch et al. PNAS | May 16, 2017 | vol. 114 | no. 20 | 5263 (21–23). In one of those studies (21), retinas from a total of three like our Pde6bH620Q/Pde6bSTOP, Pax6α::Cre mice. Our mouse aggregation chimeras expressing mutant pig rhodopsin were data suggest that in gene therapy-treated RP patients, im- shown to have 7%, 16%, or 42% normal rods distributed in provement of visual function would be stable, cone death would patches. ONL thickness in the 42% of rescued retinas was rel- be halted in treated regions, and OS shortening would only be in atively uniform and intermediate between wild-type and pre- a narrow zone bordering the treated areas. However, if our find- dominantly transgenic retinas. The second study (22) examined ings do not fully translate to human RP, it could be because our hemizygous-transgenic female mice on a retinal degenera- genetic rescue is based on allelic conversion (i.e., the “wild-type” − − tion slow (rds / ) background (transgene present on only one sequence of the PDE6b gene is restored). In contrast, the AAV- X-); in situ hybridization revealed a mosaic pattern mediated gene replacement therapy currently favored in patients of transgene expression (rescue) in retinas, and ONL thickness may not yield physiological levels of protein production—although was uniform in 2-mo-old hemizygous females. In both studies, mouse studies suggest otherwise (26). In addition, our results may the authors expected to see nonuniform ONL—rescued patches or may not apply to all types of inherited retinal degeneration, with normal thickness ONL, alternating with mutant patches of given their diverse etiologies (27). thin ONL. However, data from both of our mutants strongly Our mouse data from this study and a previous one (5) suggest that surviving photoreceptors undergo active or passive demonstrate that correction of the genetic defect is sufficient to rearrangement in response to the dramatic physical changes to prevent death in treated rods and that the extent of rescue de- the photoreceptor layer landscape—essentially filling in the void pends on the number of treated rods (more is better) and the left by degenerating rods to yield an ONL that is either uniform timing of treatment (earlier is better). in thickness (Pde6bH620Q/Pde6bSTOP, Pde6g::CreERT2 mutant) or graded (Pde6bH620Q/Pde6bSTOP, Pax6α::Cre mutant). In fact, Materials and Methods remodeling of photoreceptor cells has been observed in a canine All experiments were approved by the Institutional Animal Care and Use model of X-linked RP (XLPRA2); these diseased retinas exhibit Committee (IACUC) at Columbia University. Methods are detailed in SI Ma- patches of diseased and normal retina, which gradually disappear terials and Methods. over time (24). In addition, retinas from chimeric mice (pro- − − duced by combining morulae from rds / and wild-type mice) ACKNOWLEDGMENTS. We thank Richard Davis and members of the Jonas Stem Cell and Brown Glaucoma Laboratories for sharing ideas and equip- showed regions with normal, intermediate, and thin ONL ment. We also thank Rebecca Tuttle for critically reviewing the manuscript. thickness (25). On the other hand, the cell autonomy of rod C.A.W.-S. is funded by the German Research Council (DFG, SFB870). S.H.T. is a death may not be generalizable to all rod degenerations. member of the RD-CURE Consortium and is supported by the Tistou and Our experiments model three features of human gene therapy- Charlotte Kerstan Foundation; NIH Grants R01EY018213, R01EY024698, R01EY026682, and R21AG050437; a Research to Prevent Blindness (RPB) treated retinas. First, like gene therapy-treated retinas, our Physician-Scientist Award; the Schneeweiss Stem Cell Fund; New York State mouse retinas are mosaics of rescued and unrescued rods. Sec- (N09G-302 and N13G-275); Foundation Fighting Blindness New York Re- ond, because patients are typically diagnosed after onset of de- gional Research Center Grant C-NY05-0705-0312; Kobi and Nancy Karp; generation, we treated our Pde6bH620Q/Pde6bSTOP, Pde6g:: the Crowley Research Fund; the Gebroe Family Foundation; and Burroughs Welcome Career Awards in Biomedical Sciences Program. The Columbia Con- CreERT2 mice when roughly 30% of rods had already died. focal and animal facilities are supported by NIH Core Grants 5P30EY019007 Third, the typical single subretinal injection is restricted to a and 5P30CA013696 and unrestricted funds from RPB and Columbia small area, resulting in large swaths of untreated cells—much University.

1. Daiger SP, Sullivan LS, Bowne SJ (2013) Genes and mutations causing retinitis pig- 16. Guérin CJ, Lewis GP, Fisher SK, Anderson DH (1993) Recovery of photoreceptor outer mentosa. Clin Genet 84:132–141. segment length and analysis of membrane assembly rates in regenerating primate 2. Narayan DS, Wood JP, Chidlow G, Casson RJ (2016) A review of the mechanisms of photoreceptor outer segments. Invest Ophthalmol Vis Sci 34:175–183. cone degeneration in retinitis pigmentosa. Acta Ophthalmol 94:748–754. 17. Mitamura Y, et al. (2013) Photoreceptor impairment and restoration on optical co- 3. Shen S, Sujirakul T, Tsang SH (2014) Next-generation sequencing revealed a novel herence tomographic image. J Ophthalmol 2013:518170. mutation in the gene encoding the beta subunit of rod phosphodiesterase. 18. Beltran WA, et al. (2015) Successful arrest of photoreceptor and vision loss expands Ophthalmic Genet 35:142–150. the therapeutic window of retinal gene therapy to later stages of disease. Proc Natl 4. Davis RJ, et al. (2013) Therapeutic margins in a novel preclinical model of retinitis Acad Sci USA 112:E5844–E5853. pigmentosa. J Neurosci 33:13475–13483. 19. Sarra GM, et al. (2001) Gene replacement therapy in the retinal degeneration slow 5. Koch SF, et al. (2015) Halting progressive neurodegeneration in advanced retinitis (rds) mouse: The effect on retinal degeneration following partial transduction of the pigmentosa. J Clin Invest 125:3704–3713. retina. Hum Mol Genet 10:2353–2361. 6. Marquardt T, et al. (2001) Pax6 is required for the multipotent state of retinal pro- 20. Zhang L, et al. (2016) Reprogramming metabolism by targeting sirtuin 6 attenuates genitor cells. Cell 105:43–55. retinal degeneration. J Clin Invest 126:4659–4673. 7. Prigge JR, et al. (2013) Nuclear double-fluorescent reporter for in vivo and ex vivo 21. Huang PC, Gaitan AE, Hao Y, Petters RM, Wong F (1993) Cellular interactions impli- analyses of biological transitions in mouse nuclei. Mammalian Genome 24:389–399. cated in the mechanism of photoreceptor degeneration in transgenic mice expressing 8. Wang W, et al. (2016) Two-step reactivation of dormant cones in retinitis pigmentosa. a mutant rhodopsin gene. Proc Natl Acad Sci USA 90:8484–8488. Cell Reports 15:372–385. 22. Kedzierski W, Bok D, Travis GH (1998) Non-cell-autonomous photoreceptor de- 9. Applebury ML, et al. (2000) The murine cone photoreceptor: A single cone type ex- generation in rds mutant mice mosaic for expression of a rescue transgene. J Neurosci presses both S and M opsins with retinal spatial patterning. Neuron 27:513–523. 18:4076–4082. 10. Aït-Ali N, et al. (2015) Rod-derived cone viability factor promotes cone survival by 23. Rattner A, Sun H, Nathans J (1999) Molecular genetics of human retinal disease. Annu stimulating aerobic glycolysis. Cell 161:817–832. Rev Genet 33:89–131. 11. Xiong W, MacColl Garfinkel AE, Li Y, Benowitz LI, Cepko CL (2015) NRF2 promotes 24. Beltran WA, Acland GM, Aguirre GD (2009) Age-dependent disease expression de- neuronal survival in neurodegeneration and acute nerve damage. J Clin Invest 125: termines remodeling of the retinal mosaic in carriers of RPGR exon ORF15 mutations. 1433–1445. Invest Ophthalmol Vis Sci 50:3985–3995. 12. Clarke G, et al. (2000) A one-hit model of cell death in inherited neuronal degener- 25. Sanyal S, Zeilmaker GH (1984) Development and degeneration of retina in rds mutant ations. Nature 406:195–199. mice: Light and electron microscopic observations in experimental chimaeras. Exp Eye 13. Young RW (1976) Visual cells and the concept of renewal. Invest Ophthalmol Vis Sci Res 39:231–246. 15:700–725. 26. Nishiguchi KM, et al. (2015) Gene therapy restores vision in rd1 mice after removal of 14. Clarke G, et al. (2000) Rom-1 is required for rod photoreceptor viability and the a confounding mutation in Gpr179. Nat Commun 6:6006. regulation of disk morphogenesis. Nat Genet 25:67–73. 27. Hurley JB, Chao JR (2015) It’s never too late to save a photoreceptor. J Clin Invest 125: 15. Rapp LM, Fisher PL, Dhindsa HS (1994) Reduced rate of rod outer segment disk syn- 3424–3426. thesis in photoreceptor cells recovering from UVA light damage. Invest Ophthalmol 28. Guo L, et al. (2014) Direct optic nerve sheath (DONS) application of Schwann cells Vis Sci 35:3540–3548. prolongs retinal ganglion cell survival in vivo. Cell Death Dis 5:e1460.

5264 | www.pnas.org/cgi/doi/10.1073/pnas.1615394114 Koch et al.